Thank you for your question!

Failures can occur from engineered, financial, biological systems to most other entities the are at risk. The traditional way to assess failure is through reliability analysis: the distribution of consequences given an event, couples probabilities with magnitude of the consequences. We can extend the idea to developing probabilistic trees (many branches connecting decisions to outcomes) ordered sequentially from the initial decision to the many outcomes. Arcs between them are probabilistic. Bayesian networks are a way to represent a hierarchy of events, their connections, the conditioning (dependencies), and so on. You can also use deterministic networks to understand cause and effect, before worrying about the probability of failure. In my opinion, describing the way events occur, leads to failure or fail for internal reasons is the critical initial step to failure analysis. Here you might think of fault trees and event tress: logical description on how things can go bad, and the ramification of their consequences. So, we have: deterministic representations, probability distributions (consequences or outcomes, frequencies given an event such as a failure), Bayesian networks, and fault-trees, event trees (these are logical maps and can include probability of failure. NASA makes available reports dealing with fault and event trees. There are many other ways; multi-objective decision analysis methods may be important to your work. Finally, keep in mind that many hazardous conditions generate routine and extreme events: their analysis may be conceptually the same but the probabilistic models are less commonly known e.g., fat tailed distributions such as the extended Pareto distribution). So extreme events analysis can be another area of your interest.

Rather than bore you, let me stop here. The book by Yacov Haimes, Risk modeling, assessment and management, 3rd edition, Wiley (2009) is a very good and very extensive reference for many of the available techniques. You might try to get it through an interlibrary loan.

All the best and thanks again.

Paolo

Hello Oussama!

I advise you to use dynamic Bayesian networks. I think it is a very suitable method for dynamic systems.

Good luck

Hi,

I send some options with a short comment.

Statistical Process Control: Easy in routine application for the determination of stability and predictability of a process

Zero Failure Principle: The goal is to minimize undesirable failures and accidents

Design of Experiment: Cause-and-effect relationship (managing inputs and optimizing outputs)

Fault Tree Analysis: Deductive analysis (diagram to identify failure causes)

Failure Mode and Effect Analysis: Potential failure impact; detection and prevention measures; responsibility and scheduled improvement

Progressive Worst Case Scenario Analysis: Impact through extreme conditions

Quality Function Deployment: Emphasis to customer satisfaction and improvement of product design – “Positive Quality”

Hazard Analysis and Critical Control Points: Threat analysis, critical point control and monitoring, preventive and corrective actions, and checking efficiency and system records

LEAN: Emphasis to speed, efficiency and removing waste (identify customers, map the value stream, create flow, establish pull approach and seek continuous improvement)

The back anlysis in big works when they happen is very usefull when we are analyzing ctastrophic succes.